TY - GEN
T1 - Adaptive Dispersion Network of Multiple Drones Based on Reinforcement Learning
AU - Maikuma, Ryota
AU - Kawai, Tenta
AU - Bando, Mai
AU - Hokamoto, Shinji
N1 - Publisher Copyright:
© 2024 IEEE.
PY - 2024
Y1 - 2024
N2 - This study discusses how to construct an observation network consisting of drones in three-dimensional space. In the network, each drone adaptively changes its motion to disperse from other drones considering relative positions to others. Such movement is independently decided under a distributed control strategy without any communication to others. Since the environment perceived by a drone changes according to other drones' motion, reinforcement learning is used to generate an adaptive action of drones. Besides, to avoid the curse of dimensionality in reinforcement learning process, this study utilizes a policy-based strategy for the states and actions. This paper explains several key factors (state, reward, etc.) in the network construction problem. Some typical results of numerical simulations are shown and indicate the effectiveness of the key factors/setting in the learning process. Moreover, it is shown that the obtained variables in the learning process indicate effective actions for the dispersion network when the number of drones changes in a mission.
AB - This study discusses how to construct an observation network consisting of drones in three-dimensional space. In the network, each drone adaptively changes its motion to disperse from other drones considering relative positions to others. Such movement is independently decided under a distributed control strategy without any communication to others. Since the environment perceived by a drone changes according to other drones' motion, reinforcement learning is used to generate an adaptive action of drones. Besides, to avoid the curse of dimensionality in reinforcement learning process, this study utilizes a policy-based strategy for the states and actions. This paper explains several key factors (state, reward, etc.) in the network construction problem. Some typical results of numerical simulations are shown and indicate the effectiveness of the key factors/setting in the learning process. Moreover, it is shown that the obtained variables in the learning process indicate effective actions for the dispersion network when the number of drones changes in a mission.
UR - http://www.scopus.com/inward/record.url?scp=85208265580&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85208265580&partnerID=8YFLogxK
U2 - 10.1109/CoDIT62066.2024.10708151
DO - 10.1109/CoDIT62066.2024.10708151
M3 - Conference contribution
AN - SCOPUS:85208265580
T3 - 10th 2024 International Conference on Control, Decision and Information Technologies, CoDIT 2024
SP - 223
EP - 228
BT - 10th 2024 International Conference on Control, Decision and Information Technologies, CoDIT 2024
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 10th International Conference on Control, Decision and Information Technologies, CoDIT 2024
Y2 - 1 July 2024 through 4 July 2024
ER -